The simulation returns a predicted tag profile in the direction
perpendicular to image tag lines as a function of
time and is normalized to lie between zero and one.
To create an energy field, the following approach is
taken. For points within tag lines in the
the vertical orientation, a set of profiles
are concatenated along the vertical axis to create
a correlation kernel, . The correlation kernel is
then successively rotated to create kernels along
other orientations. Additional energy fields are
constructed using correlation kernels for tag
endpoints.
Let the image be represented by
. The normalized correlation,
is
with , and
with
when
is a constant multiple of
.
In order to increase the discrimination power of the technique
the energy field is set to
, where
is a
positive integer less than 10. Endpoint
energy fields are termed
, and are obtained
from correlating endpoint masks with the tag data.
For each tag endpoint, a correlation mask is generated,
and is used to construct endpoint energy images for
subsequent frames.
These kernels essentially have half of the window
filled with tag profiles, and depending on the tag line,
have half filled with
zero intensities, or intensities from surrounding
organs.
Currently, tag endpoint coordinates are specified
in the first frame. All subsequent endpoints are
determined by the algorithm.